Beruflich Dokumente
Kultur Dokumente
1982
AIAA 82-0070R
T HIS report reviews the methods of Kalman filtering in significant uncertainties in the modeling. For autonomous
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
attitude estimation and their development over the last spacecraft the use of inertial reference units as a model
two decades. This review is not intended to be complete but is replacement permits the circumvention of these problems. In
limited to algorithms suitable for spacecraft equipped with this representation the angular velocity of the spacecraft is
three-axis gyros as well as attitude sensors. These are the obtained from the gyro data. The kinematic equations are
systems to which we feel that Kalman filtering is most ap- used to obtain the attitude state and this is augmented by
plicable. means of additional state-vector components for the gyro
The Kalman filter uses a dynamical model for the time biases. Thus, gyro data are not treated as observations and the
development of the system and a model of the sensor gyro noise appears as state noise rather than as observation
measurements to obtain the most accurate estimate possible noise.
of the system state using a linear estimator based on present It is theoretically possible that a spacecraft is three-axis
and past measurements. It is, thus, ideally suited to both stabilized with such rigidity that the time development of the
ground-based and on-board attitude determination. However, system can be described accurately without gyro information,
the applicability of the Kalman filtering technique rests on the or that it is one-axis stabilized so that only a single gyro is
availability of an accurate dynamical model. needed to provide information on the time history of the
The dynamic equations for the spacecraft attitude pose system. The modification of the algorithms presented here in
many difficulties in the filter modeling. In particular, the order to apply to those cases is slight. However, this is of little
external torques and the distribution of momentum internally practical importance because a control system capable of such
Eugene J. Lefferts was born in 1925 in New York City and grew up in Philadelphia. He received the B.S. in Education from Westchester State
College, Westchester, Pennsylvania in 1951 and the M.A. in Mathematics from Temple University in Philadelphia in 1953. From 1953 until 1967 he
was at the Glenn L. Martin Company (now Martin-Marietta Corporation), where he worked on problems of analog simulation, optimal control,
filtering theory, and Lyapunov stability theory. From 1959 until 1963 he was a member of the RIAS group at Martin. Since 1967 he has been with
NASA Goddard Space Flight Center in Greenbelt, Maryland and since 1972 he has been head of the Attitude Dynamics Section of the Mission and
Data Operations Division at GSFC. He has written numerous articles on estimation, control, and stability and has been responsible for attitude
mission analysis for numerous NASA missions including recently: Applied Technology Satellite 6 (ATS-6), Solar Maximum Mission, and
Landsat-D.
F. Landis Markely was born in 1939 in Philadelphia, Pennsylvania and attended public schools in Cheltenham Township, Pennsylvania. He
received a Bachelor of Engineering Physics from Cornell University in 1962 and the Ph. D in Physics from the University of California at Berkley in
1967. From 1967 to 1968 he was a National Science Foundation Post Doctoral Fellow at the center for Theoretical Physics at the University of
Maryland and from 1968 to 1974 he was assistant Professor of Physics at Williams College, Williamstown, Massachusetts. In 1974 he joined the
Systems Sciences Division of the Computer Sciences Corporation in Silver Spring Maryland, where he was active in mission analysis in software
development for several NASA missions, among them, NIMBUS and the Solar Maximum mission, and also in general spacecraft dynamics and
simulation studies. Since 1978 Dr. Markely has been a physicist in the Aerospace Systems Division of the Naval Research Laboartory, where he has
continued to work extensively in the areas of spacecraft attitude and orbit dynamics, estimation and control. Dr. Markely is the author of numerous
journal articles and conference reports in theoretical physics and in Aerospace and Astrodynamic studies. He was also one of the major con-
tributors to the book, "Spacecraft Attitude Determination and Control," edited by J.R. Wertz.
Malcolm D. Shuster was born in 1943 in Boston, Massachusetts and attended public schools in greater Boston. He receive the S.B. in Physics
from the Massachusetts Institute of Technology, the Ph.D. in Physics from the University of Maryland, and a M.S. in Electrical Engineering from
the Johns Hopkins University. Between 1970 and 1977 he pursued an academic career in theoretical physics during which he held positions with the
French Atomic Energy Commission, Paris, France (engineer-physicist, 1970-72), the University of Karlsruhe, West Germany (instructor, 1972-73),
Tel-Aviv University, Israel (lecturer, 1973-76) and Carnegie-Mellon University, Pittsburgh, Pennsylvania, (visiting assistant professor, 1976-77).
During this period he carried our research on the interaction of elementary particles with nuclei and taught and lectured extensively in English,
French and German. In 1977 he joined the System Sciences Division of the Computer Sciences Corporation in Silver Spring, Maryland, where he
provided mission analysis and software support for a number of NASA missions including: Seasat, Magsat, Applications Explorer Mission, Solar
Maximum Mission, Dynamics Explorer Mission, and the Space Telescope. Since 1981 he has been with Business and Technological Systems, Inc.,
where he works on theoretical and practical problems of estimation, dynamics, and control for space and defense systems. Dr. Shuster has written
numerous articles on theoretical physics, estimation, and control and is currently preparing a book on spacecraft attitude estimation.
Presented as Paper 82-0070 at the AIAA 20th Aerospace Sciences Meeting, Orlando, Fla., Jan. 11-14, 1982; submitted Dec. 11, 1981; revision
received May 21, 1982. This paper is declared a work of the U.S. Government and therefore is in the public domain.
417
418 LEFFERTS, MARKLEY, AND SHUSTER J.
accuracy invariably includes gyros in the control loop and earliest published reference is by Farrell,11'12 who studied the
these may be sampled to obtain the full angular velocity. ; extent to which Kalman filtering of crude attitude
Therefore, with little loss of generality, the algorithms here measurements from sun sensors and magnetometers could
are specialized to spacecraft with three-axis gyros. provide attitude accuracy equivalent to that obtained without
Throughout this report the attitude is represented by the smoothing from more elaborate instrumentation. Farrell
quaternion. The development of the Kalman filter for the represented the attitude by Euler angles and assumed torque-
quaternion representation was motivated by the requirement free motion in the attitude prediction. Cherry and O'Con-
of realrtime autonomous attitude determination for attitude nor,13 iii their design of the lunar excursion module autopilot,
control and the annotation of science data. The quaternion considered sequential estimation of the disturbance torques
parameterization was chosen for several practical reasons: 1) induced by the ascent or descent propulsion system. Potter
the prediction equations are treated linearly, 2) the and Vander Velde14 used Kalman filtering theory to determine
representation is free from singularities (thus the gimbal-lock the optimum mixing of gyroscope and star tracker data in an
situation is avoided), and 3) the attitude matrix is algebraic in attitude determination system. Generally, as remarked by
the quaternion components (thus eliminating the need for Sabroff,15 the application of. Kalman filtering to attitude
transcendental functions). estimation had not shown impressive results up to 1967. Aside
The use of the quaternion as the attitude state presents some from insufficient study, the lack of real success in applying
difficulty in the application of the filter equations. This optimal estimation was caused by the inability to model the
difficulty is due to the lack of independence of the four system dynamics accurately.
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
quaternion components, which are related by the constraint Continued effort in this area was evidenced at a symposium
that the quaternidn have unit norm. This constraint results in on spacecraft attitude determination in 1969, at which six
the singularity of the covariance matrix of the quaternion papers were presented on the application of Kalman filtering
state. The various ways to treat or circumvent this difficulty to this problem. Foudriat16 and Arneson and Nelson17
make up the major part of this report. Each of the filters considered spin-stabilized satellites, while Ribarich18 and
discussed predicts the quaternion state in the same manner as Lesinski19 were interested in the dual-spin case. Pauling,
if each component were independent. Different approaches to Jackson, and Brown20 and Toda, Heiss, and Schlee21 studied
the update equations and to the covariance representation and the space precision attitude reference system (SPARS), which
its propagation give rise to the different methods presented. used gyro measurements in a model replacement mode with
In principal, one can also estimate other sensor biases by periodic star-sensor updates. Pauling et al.20 used the attitude
suitably augmenting the state vector. In general, however, matrix for state prediction and Euler angles for updates, while
these other biases are not estimated simultaneously with the Toda et al.21 used the quaternion for prediction and in-
attitude due to problems of observability under normal cremental error angles in the update. Jackson22 also con-
spacecraft operating conditions. Therefore, the estimation tributed a paper to the symposium on the application of
methods for sensor biases other than gyro drift biases will not nonlinear estimation theory to the attitude determination
be treated here. problem, an application that was also studied by Kau, Kumar,
Section II of this report reviews the relevant literature on andGranley.23
Kalman filtering as applied to attitude estimation and on In a 1971 review of strapdown navigation by Edwards,24
related topics. Since the attitude problem is nonlinear, the Kalman filtering is not treated but the use of the quaternion
vehicle for optimal estimation is the extended Kalman filter, and error angles for state prediction is discussed. A review
which is reviewed in Sec. Ill without derivation. Attitude article by Shmidtbauer, Samuelsson, and Carlsson25 extends
kinematics are reviewed in Sec. IV, with emphasis on the the history to 1973 and contains a classification and
quaternion representation. A discussion of gyros used in the discussion of extended Kalman filter methods for attitude
model replacement mode and the modeling of attitude sensors determination, with specific emphasis on algorithms suitable
is presented in Sec. V. for bn-board computation.
The state equation and the equation for attitude prediction
are derived in Sec. VI. Various approaches to the filtering
equations are discussed in the succeeding five sections. Attitude Representations
Section XII reviews the advantages of the different methods. The nonlinear aspects of rotational kinematics are
Several results of interest have been gathered in the Appendix. discussed in many texts.26-30 Approaches which treat the
The intent of the present work is to provide a complete spacecraft axes as uncoupled (thus ignoring commutation
account of what now seem to be the relevant filtering errors) have been studied by Potter and Vander Velde,14
algorithms rather than to present a new approach. An attempt Ribarich,18 Schmidtbauer et al.,25 and Farrenkorif.31 This
has been made to present the current methods in a common approximation leads to independent linear estimation
framework in order to produce a useful reference for future problems for the three spacecraft axes* and closed-form
applications. expressions for the steady-state estimation errors can be
obtained in some cases.14'25'31 Uncoupled-axis Kalman filters
employing both gyro data and dynamics modeling have been
II. Historical Survey used in on-board attitude control systems, in the NASA
Early Applications of Kalman Filtering International Ultraviolet Explorer and Solar 'Maximum
The Kalman filter,1-2 which was originally developed as a Mission32 spacecraft, for example.
tool in linear estimation theory, was soon applied to nonlinear Stuelpnagel33 and Markley34 discuss different para-
orbital guidance and navigation problems in the Apollo meterizations of the attitude that have been used when the
program by Schmidt and his collaborators.3'5 Almost decoupled-axis approximation, is not applicable. Early
simultaneously with Kalman's work, Swerling6'7 developed a strapdown navigation systems used the direction-cosine
recursive theory for satellite navigation which differed from matrix as the representation of the attitude.20'24 Due to round-
Kalman's in the treatment of the process noise and was not off, quantization, and truncation errors in the attitude
carried out as completely. Standard treatments of8 Kalman propagation, this procedure results in an attitude matrix that
filtering can be found in the review of Sorenson and the is not orthogonal.35'36 Various orthogonalization schemes for
textbooks of Jazwinski9 and Gelb.10 the attitude matrix were developed; but Giardina, Bronson,
An accurate and complete historical survey of Kalman and Wallen36 proved that an optimal orthogonalization (one
filtering for attitude estimation is not possible since the that minimizes the sum of squares of the differences between
earliest applications were directed toward national defense the elements of the propagated matrix and those of the or-
and hence could not be published in the open literature. The thogonalized matrix) requires a computationally expensive
SEPT.-OCT. 1982 KALMAN FILTERING FOR SPACECRAFT ATTITUDE ESTIMATION 419
matrix square root. Because of this problem and the of gyro noise models in Kalman filters were generally in-
redundancy of the nine-parameter direction cosjne complete. Thus, Potter and Vander Velde14 include only the
representation, it has not been widely used recently. random walk term in the drift rate, while both Pauling et al.20
The three-parameter Euler-angle representation26'30'33'34 and Toda et al.21 have only the white noise term.
was used in several early applications of Kalman filtering to Farrenkopf31'69 considered a gyro model including all the
attitude estimation.11'16'17'19'20'23 However the kinematic terms discussed above. This model was used in the subsequent
equations for Euler angles involve nonlinear and com- HEAO system development.56'57
putationally expensive trigonometric functions, and the
angles become undefined for some rotations (the gimbal-lock III. The Kalman Filter
situation), which cause problems in Kalman filtering ap- We review in this section the principal equations for the
plications. Despite these difficulties, the Euler angle extended Kalman filter9'10 in order to introduce the necessary
representation continues to be used for the attitude estimation notation for the sections which follow.
of spinning spacecraft. Stuelpnagel33 discusses two other The state equation may be written as
three-parameter representations: the exponential of a 3 x 3
skew-symmetric matrix (rotation-vector representation) and
the Cayley parameterization (not to be confused with the x(t)=f(x(t),t)+g(x(t),t)w(t) (1)
at
Cayley-Klein parameters). The latter is equivalent to the
Gibbs-vector parameterization.34 Neither of these repre-
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
sentations has found a Kalman filtering application, to where x ( t ) is the state vector and w ( t ) , the process noise, is a
our knowledge. Stuelpnagel proves that no three-parameter Gaussian white-noise process whose mean and covariance
representation can be both global and nonsingular. function are given by
The global nonsingular four-parameter representation of
the attitude in terms of Euler's symmetric parameters, or E[w(t)]=0 (2)
equivalently the four components of a quaternion, is E[w(t)wT(t')]=Q(t)5(t-t')
discussed by many authors.26'30'33'37 Quaternions were in- (3)
vented by Hamilton38 in 1843; their use in attitude dynamics E denotes the expectation and T the matrix transpose. The
simulations was promoted by Robinson39 and by Mitchell and initial mean and covariance of the state vector are given by
Rogers.40 The attitude matrix computed from a quaternion
(as a homogeneous quadratic function) is orthogonal if the (4)
sum of squares of the quaternion components is unity. If E[x(t0)}mx(t0)=x0
propagation errors result in a violation of this constraint, the
quaternion can be renormalized by dividing its components by E( [x(t0) -x0] (x(t0) - X O ] T ] m =P (5)
the (scalar) square root of the sum of their squares; and Prediction
Giardina et al.36 showed that the attitude matrix computed
from the renormalized quaternion is identical to the one given Given the initial conditions on the state vector and the state
by their optimal orthogonalization. The application of covariance matrix, the minimum variance estimate o^the state
quaternions to strapdown guidance, with error analyses, was vector at a future time t is given in the absence of
discussed by Wilcox35 and Mortenson.41 Quaternion measurements by the conditional expectation
kinematics has been the subject of several recent studies.42'48
Of particular concern has been the best method for extracting x(t)=E[x(t) \x(t0)=x0] (6)
a quaternion from a attitude matrix.49'52 A recent review of
quaternion relations has been given by Friedland.53 This predicted estimate satisfies the differential equation
The advantages of the quaternion parameterization have
led to its frequent use in attitude determination systems. One - x ( t ) =E(f(x(t),t) } mf(x(t),t) (7)
example is an attempt by Lefferts and Markley54 to model the at
attitude dynamics of the NIMBUS-6 spacecraft, which in-
dicated that dynamics modeling with elaborate torque models which we write approximately as
could still not give acceptable attitude determination ac-
curacy. For this reason, most applications of quaternion
attitude estimation have used gyros in the dynamic model (8)
at
replacement mode. These include the work by Toda et al.21
mentioned above, by the group at TRW for the precision Equation (8) may be integrated formally to give
attitude determination system (PADS),55 which was in-
corporated in the attitude determination system for the High- + (t,x(t0),t0) (9)
Energy Astronomy Observatory (HEAO) mission,56'58 by
Yong and Headley59 for highly maneuverable spacecraft, by The state error vector and covariance matrix are defined by
Murrell60 in a design for the NASA Multimission Modular
Spacecraft, by Sorenson, Schmidt, and Goka61 in a square- Ax(t)=x(t)-x(t) (10)
root filtering application, by Shuster and co-workers62'64 for a
microprocessor-based on-board attitude determination P(t)=E[Ax(t)AxT(t)] (11)
system, and by Markley65 for an autonomous navigation
study. Neglecting terms which are higher than first order in the state
error vector and the process noise, the state error vector
Gyro Noise Models satisfies the differential equation
The first papers describing statistical models of gyro drifts d_
were by Newton66 and Hammon67 in 1960. Newton con- (12)
sidered additive white noise in the gyro drifts, while Hammon ~dt
assumed that the gyro drift rates were an exponentially where
correlated random process. Dushman68 considered a drift-
rate model obtained by adding a random walk component to F(t)mf(x,t) (13)
Hammon's autocorrelation function. Early implementations x(t)
420 LEFFERTS, MARKLEY, AND SHUSTER J. GUIDANCE
Equation (12) may be integrated formally to give Equation (29) is the starting point for many efficient and
numerically stable factored forms of the Kalman filter.70
U')G(f')w(f')d/' (15) In the sections which follow, the implementation of the
above equations is examined for three different represen-
tations of the attitude estimation problem. In each case,
where $(t,t0) is the transition matrix, which satisfies explicit expressions are developed for the transition matrix $
and the sensitivity matrix H.
A
dt-
IV. Attitude Kinematics
(17) In the systems investigated in the present study the attitude
is represented by the quaternion defined as
Note that for nonlinear systems $(t,t0) also depends im-
plicitly on x(t0), which for notational convenience will be
suppressed. (30)
The predicted covariance matrix satisfies the Riccati
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
equation where
q4=cos(0/2) (3D
with The vector b is the drift-rate bias and if, is the drift-rate noise.
ij; is assumed to be a Gaussian white-noise process
E[n,(t)}=0 (49)
(39) E(i,(t)r,T(t')]=Ql(t)S(t-t') (50)
E(ri2(t)ril(t')]=Q2(t)d(t-t') (53)
d_
(41)
~di The two noise processes are assumed to be uncorrelated
0 (55)
s'
It will be convenient in later sections to define the four- gives rise to an exponentially correlated noise term as in the
component quantity model considered by Hammon.67 In order to reproduce
realistic gyro data, a superposition of several drift terms, with
different values of the time constant may be needed.68 Since
(43) relatively persistent drifts are observed in actual gyro
operation, at least one time constant must be very large.
whence Letting r be infinite, which leads to Eq. (51), is adequate for
most applications.
d^ In the propagation equations to be used in the Kalman
(44) filter, Eq. (48) is assumed to be integrated continuously. The
~dt model thus assumes that the gyros are used in a rate mode. In
practice, however, rate-integrating gyros are used, which
If the direction of w is constant over the time interval of sense the spacecraft angular rates continuously but are
interest or if the *'rotation vector" defined by sampled at discrete intervals. The spacecraft attitude is also
propagated at discrete intervals, equal to either the gyro
A0= f u(t')dtf (45) sampling intergal or some multiple thereof. If the attitude
update interval is much shorter than the Kalman filter update
interval, as it always is in practice, the approximation of
is small, then the solution of Eq. (41) is continuous gyro updates will be good.
(46) Attitude Sensors
The attitude sensors considered here may be any sensor for
where which the measured quantity depends solely on the direction
of some object in the sensor coordinate system. Thus, we treat
sin(!A0l/2) the most general case possible, although in most practical
M(A0)=cos(lA0l/2)/, x ,+ (47)
IA0I applications of Kalman filters for attitude estimation, these
sensors will be vector sun sensors and star trackers, which are
V. Sensor Models the most accurate.
The direction of a body in the sensor coordinate system ps
The spacecraft being studied is assumed to be equipped with is related to the direction in the reference coordinate system
three-axis gyros as well as (nongyroscopic) attitude sensors. PR according to
Gyro Models = TA(q)pR (56)
We use a simple but realistic model for gyro operation
developed by Farrenkopf31 and applied to the HEAO mission where A(q) is the spacecraft attitude and T the sensor
by Hoffman and McElroy.57 In this model the spacecraft alignment matrix. Note that the measurement depends ex-
angular velocity is related to the gyro output vector u ac- plicitly on the attitude but not on parameters such as gyro
cording to biases.
It will be assumed throughout this report that the sensor
(48) measurements are scalar and uncorrelated. In general, the
422 LEFFERTS, MARKLEY, AND SHUSTER J. GUIDANCE
(58)
Prediction
The seven-dimensional state error vector satisfies the
(59) differential equation
d_
Noting that the matrix function 0 is linear and homogeneous + G(t)w(t) (69)
in its argument and defining the 4 x 3 function S(q) by dt
(60) with
q2 (72)
Q,(t)
6(0 = (73)'
Q2(t)
The properties of the matrix S(q) are discussed in the Ap-
pendix. F(t) and G(f) are determined straightforwardly from Eqs.
(13), (14), (59), and (61).
Prediction The transition matrix has the general form
The predicted state vector is defined as in Sec. II. Taking Q(t,t0)
the expectation of Eqs. (59) and (61) leads, within the ap- (74)
proximation of Eq. (8), to 0'3x4 I
3x.3
discussed in Sec. V, it follows that the sensitivity matrix is a It is also shown in the Appendix that
seven-dimensional row vector of the form
t0) (88)
=[t, 0T] (78)
Substituting these expressions into the expression for the
where transition matrix gives
=/_^^Ps\ I *(t,t0)=S(q(t))*(t,t0)ST(q(t0))
(79)
\dp dq ) Ift-)
| K(t,t0)
where (91)
0'3x3 ' ^3x3
pB=A(q(-))pR (82)
The second term of the right member of Eq. (89) annihilates
VIII. Singularity of the Covariance Matrix P(t0) on the left and P(t) on the right. Thus, if the 6x6
The covariance matrix for the seven-dimensional state covariance matrix P(t) is defined by
vector is singular. This follows immediately from the con-
straint on the quaternion norm so that P(t)=ST(q(t))P(t)S(q(t)) (92)
q(t) (93)
0 'o
Filtering
This state error vector cannot be treated as simply as the Analogously to Eqs. (78) and (79) in Sec. VII the sensitivity
seven-dimensional state error vector of Sec. VIII. In Sec. matrix is given by
VIII, the four components of the quaternion were treated as
independent variables with the normalization constraint being (H3)
maintained by the form of the equation of motion. In the
present case the fourth component of the quaternion error where the three-dimensional row vector ty is given by
must be determined from the constraint according to
(114)
(102)
which reduces to
Thus, partial differentiation must be treated differently in the
two cases. (115)
(',)/=<-
Prediction and is given by Eq. (81).
The truncated covariance matrix is defined as In implementing the Kalman filter in this case also the 7 x 7
covariance matrix need never be computed. The state vector is
(103) updated by computing
which satisfies
(116)
Py(t)=*y(t,t0)Py(t0)$T(t,t0)
+ (' $y(t,t')Gy( ,t')dt' (104) where Ky is computed from Py( ) and Hy using Eq. (26).
J/n
A<7,/( + ) is obtained from Eq, (102) and the updated state
vector is given by
where
(105) (117)
contained in the three vector components. Therefore, the six- it follows immediately that
component object defined by the vector components of the
incremental quaternion and the drift-bias vector will provide a 6# = */2 [
nonredundant representation of the state error. This d?
representation will turn out to be identical to that developed in
Sec. IX. (131)
If the infinitesimal attitude error angles are defined as twice
the vector components of the incremental quaternion, then where
this treatment becomes very similar to that employed by
several other authors.21'59'65 6(0 <o<o
(132)
Define the error quaternion as 0 o
(118) Now
and the six-dimension body-referenced state vector as (133)
0
dq
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
and d
&x(t)=F(t)Ax+G(t)w(t) (137)
at
H T (<?) ! 03X} 1 f q -]
11
X= (123) where
0.x, 1 /3XJ J L ft J
_ O
-
\ Lf j A, (124) ^(0 ="--! ------- (138
L 3X3 ! 03X3 J
i then follows that
T <> 1 fi( f\
r -;/2/5x5 '1 o3x3 i mo^
(125)
^3x3 i ^3x3
*3
x*S(q)x (128)
(141)
The results of Sec. IX may now be derived in short order.
*(t,t (142)
ot
Prediction
From subject to
d (143)
q (129)
at
It now follows immediately that
-~q (130) (144)
at
426 LEFFERTS, MARKLEY, AND SHUSTER J. GUIDANCE
with
so that
A(t,t0) ! (161)
(146)
3x3 3x5
Noting from Eqs. (37) and (60) that
as in Eq. (91).
The covariance matrix is defined as (162)
(1'47) (163)
and from Eq. (127) it follows that we see that this algorithm is identical to implementing Eq.
(100).
P(t)=ST(q(t))P(t)S(q(t)) (148) We note again that the quantities AJC( + ), 64 ( + ).,. and
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
The 6x6 body representation of the covariance matrix Substituting this expression into Eq. (A10) above, recalling
presented in Sees. IX and XI preserves the proper rank of the Eq. (A7), and noting that c(t) is arbitrary leads directly to
covariance matrix while simplifying the computation con- Eq.(Al).
siderably. The transition matrix and the covariance matrix are To prove Eq. (A2) examine the quantity
computed as 6 x 6 matrices throughout. Of particular value in
this representation is the fact that the elements of the C(t,t0) =Q(t,t0)3,(q(t0)) -Z(q(t))h(t,t0) (A12)
covariance matrix have a simple interpretation in terms of
gyro bias errors and angular errors in the body frame. In Differentiating this expression and using Eq. (Al) leads to
addition, the matrices G and H have a particularly simple
form. d_
In any Kalman filter implementation the largest com- dt . '
putational burden is imposed in general by the computation
of the transition matrix and the contribution of the process which may be integrated directly to yield
noise to the state covariance matrix. These quantities are not
needed at the same level of accuracy as the state vector and C(t,t0)=Q(t,t0)C(t0,t0) (A14)
hence may be computed at much larger intervals than is
required by the state vector propagation. For this same But from Eq.(A12)
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
10 36
Gelb, A., Ed., Applied Optimal Estimation, MIT Press, Cam- Giardina, C.R., Bronson, R., and Wallen, L., "An Optimal
bridge, Mass., 1974. Normalization Scheme," IEEE Transactions on Aerospace and
n
Farrell, J.L., "Attitude Determination by Kalman Filtering," Electonic Systems, Vol. AES-11, July 1975, pp. 443-446.
37
Vol. 1, NASA-CR-598, Sept. 1964. Fallon, L., "Quaternions," Spacecraft Attitude Determination
12
Farrell, J.L., "Attitude Determination by Kalman Filtering," and Control, edited by J.R. Wertz, D. Reidel, Dordrecht, the
Automatica, Vol. 6,1970, pp. 419-430. Netherlands, 1978.
13 38
Cherry, G.W. and O'Connor, J., "Design Principles of the Hamilton, W.R., Elements of Quaternions, Longmans, Green,
Lunar Excursion Module Autopilot,'' MIT Rept. R-499, July 1965. and39Co., London, 1866.
14
Potter, J.E. and Vander Velde, W.E., "Optimum Mixing of Robinson, A.C., "On the Use of Quaternions in the Simulation
Gyroscope and Star Tracker Data," Journal of Spacecraft and of Rigid-Body Motion," Wright Air Development Center, Tech.
Rockets, Vol. 5, May 1968, pp. 536-540. Rept. 58-17,^Dec. 1958.
15
Sabroff, A.E., "Advanced Spacecraft Stabilization and Control ^Mitcheli, E.E.L. and Rogers, A.E., "Quaternion Parameters in
Techniques," Journal of Spacecraft and Rockets, Vol. 5, Dec. 1968, the Simulation of a Spinning Rigid Body," Simulation, June 1965,
pp. 1377-1393; pp. 390-396.
16 41
Foudriat, E.G., "A Limited Memory Attitude Determination Mortenson, R.E., "Strapdown Guidance Error Analysis," IEEE
System Using Simplified Equations of Motion," Proceedings of the Transactions on Aerospace and Electronic Systems, Vol. AES-10,
Symposium on Spacecraft Attitude Determination, Aerospace Corp. July 1974, pp. 451-457.
42
Rept. TR-0066 (5306)-12, Vol. 1, Sept.-Oct. 1969, pp. 15-27. Ickes, B.P., "A New Method for Performing Digital Control
f7
Arneson, G.R. and Nelson, A.D., "The Development and System Attitude Computations Using Quaternions," AIAA Journal,
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
Performance of an Attitude Determination Data Reduction and Vol. 8, Jan. 1970, pp. 13-17.
43
Analysis System," Proceedings of the Symposium on Spacecraft Grubin, C., "Derivation of the Quaternion Scheme via the Euler
Attitude Determination, Aerospace Corp. Rept. TR-0066 (5306)-12, Axis and Angle," Journal of Spacecraft and Rockets, Vol. 7, Oct.
Vol. 1, Sept.-Oct. 1969, pp. 223-233. 1970, pp. 1261-1263.
18
Ribarich, J.J., "Gyrostat Precision Attitude Determination and ^Hendley, A.C., "Quaternions for Control of Space Vehicles,"
Control," Proceedings of the Symposium on Spacecraft Attitude Proceedings of the Institute of Navigation National Space Meeting of
Determination, Aerospace Corp. Rept. TR-0066 (5306)-12, Vol. 1, Space Shuttle, Space Station, and Nuclear Shuttle Navigation, George
Sept.-Oct. 1969, pp. 387-396. C. Marshall Space Flight Center, Huntsville, Ala., Feb. 1971, pp. 335-
19
Lesinski, I.E., "Attitude Determination Performance Potential 352.
45
for a Yaw-Spin Satellite, Part One: Formulation of the Estimation Altman, S.P., "A Unified State Model of Orbital Trajectory and
Equations," Proceedings of the Symposium on Spacecraft Attitude Attitude Dynamics," Celestial Mechanics, Vol. 6, Dec. 1972, pp. 425-
Determination, Aerospace Corp. Rept. TR-0066 (5306)-12, Vol. 1, 446.
Sept.-Oct. 1969, pp. 397-413. ^Grubin, C., "Attitude Determiantion for a Strapdown Inertial
20
Pauling, D.C., Jackson, D.B., and Brown, C.D., "SPARS System Using the Euler Axis/Angle and Quaternion Parameters,"
Algorithms and Simulation Results," Proceedings of the Symposium AIAA Paper 73-900, Aug. 1973.
on Spacecraft Attitude Determination, Aerospace Corp. Rept. TR- 47
Mayo, R.A., "Relative Quaternion State Transition Relation,"
0066 (5306)-12, Vol. 1, Sept.-Oct. 1969, pp. 293-317. Journal of Guidance and Control, Vol. 2, Jan.-Feb. 1979, pp. 44-48.
21 48
Toda, N.F., Heiss, J.L., and Schlee, F.H., "SPARS: The Yen, K. and Cook, G., "Improved Local Linearization
System, Algorithms, and Test Results," Proceedings of the Sym- Algorithm for Solving the Quaternion Equations," Journal of
posium on Spacecraft Attitude Determination, Aerospace Corp. Guidance and Control, Vol. 3, Sept.-Oct. 1980, pp. 468-471.
49
Rept. TR-0066 (5306)-12, Vol. 1, Sept.-Oct. 1969, pp. 361-370. Klumpp, A.R., "Singularity-Free Extraction of a Quaternion
22
Jackson, D.B., "Applications of Nonlinear Estimation Theory to from a Direction Cosine Matrix," Journal of Spacecraft and Rockets,
Spacecraft Attitude Determination Systems," Proceedings of the Vol. 13, Dec. 1976, pp. 754-755.
50
Symposium on Spacecraft Attitude Determination, Aerospace Corp. Shepperd, S.W., "Quaternion from Rotation Matrix," Journal
Rept. TR-0066 (5306)-12, Vol. 1, Sept.-Oct. 1969, pp. 89-111. of Guidance and Control, Vol. 1, May-June 1978, pp. 223-224.
f3 51
Kau, S., Kumar, K.S.P., and Granley, G.B., "Attitude Deter- Spurrier, R.A., "Comment on * Singularity-Free Extraction of a
mination via Nonlinear Filtering," IEEE Transaction on Aerospace Quaternion from a Direction Cosine Matrix'," Journal of Spacecraft
and Electronic Systems, Vol. AES-5, Nov. 1969, pp. 906-911. and Rockets, Vol. 15, July-Aug. 1978, pp. 255-256.
24 52
Edwards, A., "The State of Strapdown Inertial Guidance and Grubin, C., "Quaternion Singularity Revisited," Journal of
Navigation," Navigation, Vol. 18, Winter 1971-72, pp. 386-401. Guidance and Control, Vol. 2, May-June 1979, pp. 255-256.
25 53
Schmidtbauer, B., Samuelsson, H., and Carlsson, A., "Satellite Friedland B., "Analysis Strapdown Navigation Using Quater-
Attitude Control and Stabilization Using On-Board Computers," nions," IEEE Transactions on Aerospace and Electronic Systems,
Organization Europe'enne de Recherches Spatiales, ESRO CR-100, Vol. AES-14, Sept. 1978, pp. 764-768.
54
July 1973. Lefferts, E.J. and Markley, F.L., "Dynamics Modeling for
26
Klein, F. and Sommerfeld, A., Uber die Theorie des Kreisels, Attitude Determination," AIAA Paper 76-1910, Aug. 1976.
55
Teubner, Leipzig, pp. 1897-1910 (also Johnson Reprint, New York, Iwens, R.P. and Farrenkopf, R.L., "Performance Evaluation of
1965). a Precision Attitude Determination Scheme," AIAA Paper 71-964,
27
Whittaker, E.T., Analytical Dynamics, Cambridge University Aug. 1971.
5
Press, Cambridge, England, 1937, (also Dover, New York, 1944). *Hoffman, D.P., "HEAO Attitude Control SubsystemA
28
Hamel, G., Theoretische Mechanik, Springer-Verlag, Berlin, Multimode/Multimission Design," AIAA Paper 76-1925, Aug. 1976.
57
1949, (corrected reprint, 1967). Hoffman, D.P. and McElroy, T.T., "HEAO Attitude Reference
29
Goldstein, H., Classial Mechanics, 2nd Ed., Addison-Wesley, Design," AAS Paper 78-120, March 1978.
58
Reading, Mass., 1980. Rose, E.F. and Berkery, E.A., "On-Orbit Control System
30
Farrell, J.L., Integrated Aircraft Navigation, Academic Press, Performance of the HEAO-2 Observatory," Journal of Guidance and
New York, 1976. Control, Vol. 4, March-April 1981, pp. 148-156.
31 59
Farrenkopf, R.L., "Analytic Steady-State Accuracy Solutions Yong, K. and Headley, R.P., "Real-Time Precision Attitude
for Two Common Spacecraft Attitude Estimators," Journal of Determination System (RETPAD) for Highly Maneuverable
Guidance and Control, Vol. 1, July-Aug. 1978, pp. 282-284. Spacecrafts," AIAA Paper 78-1246, Aug. 1978.
32
Markley, F.L., "Attitude Control Algorithms for the Solar ^Murrell, J.W., "Precision Attitude Determination for
Maximum Mission," AIAA Paper 78-1247, Aug. 1978. Multimission Spacecraft," AIAA Paper 78-1248, Aug. 1978.
33 61
Stuelpnagel, J., "On the Parametrization of the Three- Sorenson, J.A., Schmidt, S.F., and Goka, T., "Application of
Dimensional Rotation Group," SIAMReview, Vol. 6, Oct. 1964, pp. Square-Root Filtering for Spacecraft Attitude Control," Journal of
422-430. Guidance and Control, Vol. 2, Sept.-Oct. 1979, pp. 426-433.
34 62
Markley, F.L., "Parameterizations of the Attitude," Spacecraft Shuster, M.D., Ray, S.N., and Gunshol, L., "Autonomous
Attitude Determination and Control, edited by J.R. Wertz, D. Reidel, Onboard Attitude Determination System Specifications and
Dordrecht, the Netherlands, 1978. Requirements," Computer Sciences Corp. Rept. CSC/TM-80/6237,
35
Wilcox, J.C., "A New Algorithm for Strapped-Down Inertial 1980.
63
Navigation," IEEE Transactions on Aerospace and Electronic Gambardella, P., Church, V., Liu, K., Rao, G., Ray, S., and
Systems, Vol. AES-3, Sept. 1967, pp. 796-802. Shuster, M., "Microprocessor-Based Autonomous Attitude Deter-
SEPT.-OCT. 1982 KALMAN FILTERING FOR SPACECRAFT ATTITUDE ESTIMATION 429
67
initiation System Design," Computer Sciences Corp., Kept. Hammon, R;L., "An Application of Random Process Theory to
CSC/TM-81/6085,1981. Gyro Drift Analysis," IRE Transactions on Aeronautical and
^Shuster, M.D., "Attitude Error Analysis Program (ATTEHP) Navigational Electronics, Vol. ANE-7, Sept. 1960, pp. 84-91.
68
Mathematical Description,*' Computer Sciences Corp., Rept. Dushman, A., "On Gyro Drift Models and Their Evaluation,'
CSC/TM-81/6012,
65
Feb. 1981. IRE Transactions on Aerospace and Navigational Electronics, Vol.
Markley, F.L., "Autonomous Satellite Navigation Using ANE-9, Dec. 1962, pp. 230-234.
69
Landmarks," AAS Paper 81-205, Aug. 1981. Farrenkopf, R.L., "Generalized Results for Precision Attitude
^Newton, G.C., "Inertial-Guidance Limitations Imposed by Reference Systems Using Gyros," AIAA Paper 74-903, Aug. 1974.
70
Fluctuation Phenomena in Gyroscopes," Proceedings of the IRE, Bierman, G.J., Factorization Methods for Discrete Sequential
Vol. 48, April 1960, pp. 520-527. Estimation, Academic Press, New York, 1977.
Downloaded by Universitats- und Landesbibliothek Dusseldorf on June 11, 2013 | http://arc.aiaa.org | DOI: 10.2514/3.56190
The thirty-six papers in this volume cover developments in satellite transponders, subsystems, antennas, high-power trans-
mission, integration and testing, launch vehicles, digital techniques, and earth station technology.
Equipment analyzed and evaluated includes traveling-wave tubes for space applications, S-band power amplifier tubes, and
complex multiplexed fillers for waveguide combinations. Antennas evaluated for communications satellites include phased-
array, mechanically despun, variable, coverage, and narrow-bandwidth types.
Stabilization systems examined include spin-stabilization, three-axis attitude control, horizon sensing, and visible light earth
sensors for satellites in equatorial orbit.
Power studies examine high-power satellite design, fabrication, and testing, and high-power television transmitters. In-
terference between satellite and terrestrial transmissions is explored. A low-cost microwave converter is proposed.
Various switching options for multiple-access and demand switching are proposed, and various low-cost, low-power ground
stations are presented.
TO ORDER WRITE: Publications Dept., AIAA, 1290 Avenue of the Americas, New York, N. Y. 10019